vMotion over Distance support with EMC VPLEX Metro

Last month at EMC World we released a storage virtualization product unlike any other. EMC VPLEX not only virtualizes back-end storage from other vendors it also allows shared storage to be presented active/active over long distances. This capability makes it possible for long distance vMotion and brings a new technical term to disaster planning, Disaster Avoidance.

Imagine you’re planning a datacenter shutdown due to a power outage or building maintenance, or perhaps that wildfire is making its way to your office location. Now you have the flexibility to be able to avoid major downtime by simply swinging your workloads to an off-site datacenter.

VMware has taken notice and has officially released a KB article detailing out full support for long distance vMotion with the EMC VPLEX product. More information on this KB article can be found at http://kb.vmware.com/kb/1021215. Perhaps you want to know more about VPLEX or even see a demo of it in action? Chad Sakac posted a great breakdown of how it works along with some videos on his website, check it out now.

Created on June 8, 2010 by Rick Scherer

Posted under Backup & Recovery, Storage, vCenter, vSphere.

This blog has 5,816 views.

Tags: , , , , ,

1 Star2 Stars3 Stars4 Stars5 Stars (4 votes, average: 3.25 out of 5)

8 Comments so far

  1. Craig
    10:22 am on June 8th, 2010

    you can also do this with Netapp Metro Cluster for quite some times ago. Similar Concept.

  2. Chris
    5:46 pm on June 8th, 2010

    Actually, the NetApp MC was an attempt to span a single VMware vSphere HA cluster across two sites. In the example above there are two separate clusters, one in each data center.

    In the current version of vSphere spanned HA clusters could experience a split-brain situation when the intersite link is unavailable, but both sides have access to shared storage.

  3. Chris B
    7:02 pm on June 8th, 2010

    @ Craig – both you and Chris are correct in a sense. I am sure that the NTAP MetroCluster and LeftHand Campus Clusters plus the Yotta Yotta technology aquired by EMC in 2006 (and others I am certain) all influenced the thinking behind the design of vPlex.

    The main NEWNESS of the vPlex solution is that all volumes are active AND reside at BOTH sites with full cache consistency across the sites. This is not the case with previous solutions on the market; volumes are active on only one of the 2 split controllers. While it is possible to do long distance vMotion with MC your data will still be point back to the original (now remote) volume. This would potentially leave you vulnerable to increased latency on IO as well as a risk of VM failure should the WAN link drop. See your own best practice documentation; it states that a Storage vMotion to a volume active on the remote node should be done at the same time as a distance vMotion.

    vPlex does not require this as the volumes are active on all controllers remote and local. This means that workloads supported by vPlex can be moved without the need to migrate the data – vMotion without the extra burden of the need for data migration.

    All of the above solutions have their merits for certain use cases. vPlex is just the next evolution, and possibly a revolution, in the chase for enabling the virtualization of global data storage.

  4. Clint
    7:26 pm on June 8th, 2010

    The most obvious piece of what was discussed here is, yes we can do long distance VMotions between heterogeneous arrays attached to different clusters. Yes, we have a new term “disaster avoidance” which can be accomplished without taking on a split-brain scenario.

    But where does this really start to make a lot of sense? EMC’s concept of the Journey to the Private Cloud. Where the end goal aligned with VCE (Virtualized Computing Environment) is to allow for a true cloud operating models where cimputing resources can be consumed in flexible and efficient ways. This doesn’t just mean I can do long distance vmotions among my data centers– it means you can utilize a 3rd party provider to provide computing resources on demand. This requires agnostic technology that allows for active-active operation to allow for long distance vmotions to a cloud (cloud-bursting).

    So in summary– its different in its purpose, and underlying technology.

  5. Stephen
    7:38 pm on June 8th, 2010

    With all due respect — Netapp MetroCluster is a somewhat dated approach, and an attempt to provide an active/active “like” storage model. In all fairness, at the time it was released it was an interesting take offering both site failover and the ability to virtualize a filer’s identity on the surving site, but it did/does by this blended traditional replication with storage impersonation.

    Incidentally, some applications can have have issues with latency of the MC’s failover semantics and if this occurs — recovery can be problematic. As well, the “failback” aspect with MC requires a fair amount of administrative intervention.

    Outside of MetroCluster is a fairly complex solution to deploy and manage — it requires you to allocate roughly 50% of the storage in the surviving site to recover the failed over filer.

    VPlex on the other hand is an entirely different approach to storage virtualization, not reliant upon traditional disaster recovery replication technology but rather it leverages advanced distributed data coherency algorithms to keep data in sync, it presents a virtual LUN to the hosts that attach to it, as it virtualizes the underlying physical storage.

    The implementation options are many — underlying storage can be made up of multiple LUNs, it can leverage concatenation or slicing techniques or it can simply encapsulate devices 1 to 1. In the case of a VPlex Metro (where storage is presented over distance and kept cache coherent) the hosts that access the VPlex LUN have read/write on both sides, this means that data created on either side is automatically synchronized between site, regardless of where the IO occurs. This is not your standard primary/secondary instance replication model, this is a true “active/active” storage.

    Mind you, the underlying storage can be any fibre channel based storage, not just EMC’s, and the use cases support virtualizing storage within a data center or across data centers.

    It not only provides next generation data mobility and transport (a major enabler for migrating virtual machines across long distances, think Cloud), but offers a new way of managing disparate pools of storage, regardless of type, speed, geometry. This allow storage admins new flexibility in how they allocate, present, provisioning, migrate to/from, and retire storage.

    So on the surface, sure MetroCluster looks like VPlex, but under the covers, its night and day.

  6. Matt L
    7:40 pm on June 8th, 2010

    I would agree with what Chris B said, most particularly that there are use-case scenarios for many of these replication and “Virtual stretch” technologies. There are differences from the storage side as well as from the virtualization software side. Think about SRM with Replication Manager as a hardware/software approach that comes close to accomplishing this, but remember vPlex is not a backup-restore technology, nor is SRM just a failover-failback tool.

    vPlex is truly an enabling technology, a game changer. With it, you can extend the range and utilization of your data center across distance, with no impact on the users. It is astounding to think of how different your virtualization strategies become using this. This truly is the goal of the “Cloud” actualized.

    The goal here is to seemlessly present a fully integrated single entity data center over distance. As well, vPlex allows for the quiesceing of data while that data is live. This is far more than just a storage vmotion.

    MC will rely on storage presented to vm’s over distance after a migration takes place. With sufficient bandwidth between sites, this may prove viable, but most IO reliant applications require the data to be housed in the same location as the virtual server.

  7. Michael
    10:07 am on June 14th, 2010

    I have one question to this: I was scanning the VPLEX articles to find high availability features accross long distances.
    In case of site diusaster when things happen without any time to prepare. Is it possible that a VPLEX reacts 100% automated to the loss of any of its sides?

    It should present the lun automatically on the other side.

    Please help

  8. craig
    9:22 am on November 9th, 2010

    how much bandwidth do you require for both location to implement vPlex?


Leave a Comment

Name (required)

Email (required)



More Blog Post

Next Post: